JP7088645B2 - Data converter - Google Patents

Description

The present invention relates to a data conversion device.

Conventionally, a technique for converting sound data including voice spoken by a person into text data has been studied. According to the technique of converting sound data into text data, it is possible to record conversations and speeches made in a certain language and convert the contents into characters. Such techniques have been applied to the automatic creation of minutes and the pre-processing of translations.

Regarding the automatic creation of minutes, the following Patent Document 1 describes an automatic minutes creation system that creates minutes by converting voice data in which voices of a plurality of speakers are encoded into text information, and to text information. The processing priority is calculated for each task from the conversion progress and the desired finish date, which is the ratio of a part of the converted voice data to the whole, and at least one specific field dictionary is calculated based on the conversion progress and the processing priority. And the minutes automatic creation system that adjusts the accuracy of the minutes by selecting the number of recorded vocabulary is described.

Patent No. 4703385

In recent years, a language model capable of converting sound data to text data with high accuracy has been provided in the form of a cloud service, and technology for converting sound data to text data has become easily available. It's getting better. Such services accept input of sound data via the Internet, generate text data based on the input sound data by the language model stored in the server, and use the obtained text data on the Internet. Send it back to the client via.

The cloud service that converts sound data to text data is highly convenient, but it may be available to any user, so it may be difficult to use if you want to convert sound data containing confidential information to text data. there were. For example, when a conversation that should not be leaked to the outside is held in a conference, it may not be recommended from the viewpoint of security to convert the recording of the conference into text data by a cloud service.

Therefore, an object of the present invention is to provide a data conversion device capable of converting sound data into text data while achieving both convenience and security.

The data conversion device according to one aspect of the present invention is specified based on a determination unit for determining whether or not a predetermined sound is included in a series of input sounds, and at least a predetermined sound among the series of sounds. For the storage unit that stores the sound data of the section to be processed and the server that generates text data based on the sound data when it is determined by the determination unit that the series of sounds contains a predetermined sound. , A transmitting unit for transmitting section sound data, and a receiving unit for receiving text data generated based on the section sound data from the server.

According to this aspect, the sound data of the specified section based on a predetermined sound is transmitted to the server, and the entire input sound data is not transmitted to the server, so that the sound that should not be leaked to the outside is produced. Even if it is input, the section to be converted into text data can be limited, and the sound data can be converted into text data while achieving both convenience and security of the cloud service.

Further, in the above embodiment, when the determination unit determines that the series of sounds contains a predetermined sound, the storage unit may use at least a part of the series of sounds input after the predetermined sound in the section. It may be stored as sound data.

According to this aspect, by storing at least a part of a series of sounds input after a predetermined sound as sound data of a section, the capacity of sound data to be stored in the storage unit can be reduced. Since it is not necessary to process whether or not a predetermined sound is included in the stored sound data, the calculation load can be reduced.

Further, in the above aspect, the storage unit stores data of a series of sounds, and at least a part of the series of sounds input after the predetermined sound is set as a section from the data of the series of sounds stored in the storage unit. Further, an extraction unit for extracting as sound data may be provided.

According to this aspect, data other than the sound data of the extracted section is stored by storing the data of the input series of sounds and extracting at least a part of the series of sounds input after the predetermined sound. Even so, you will be able to select it after the fact, send it to the server, and convert it to text.

Further, in the above embodiment, a division unit for dividing the sound data of the section into a plurality of sound data is further provided, and the transmission unit rearranges the order of the plurality of sound data and transmits the plurality of sound data to the server. The receiver receives a plurality of text data generated based on a plurality of sound data, and synthesizes the plurality of text data into one text data based on the rearrangement of the order of the plurality of sound data by the transmitter. Further units may be provided.

According to this aspect, the sound data of the section specified based on a predetermined sound is divided into a plurality of sound data, and the order is changed and transmitted to the server, so that the content of the transmitted sound data can be obtained. It is possible to prevent it from being read by a third party.

Further, in the above aspect, the transmission unit may distribute and transmit a plurality of sound data to a plurality of servers that generate text data based on the sound data.

According to this aspect, sound data in a section specified based on a predetermined sound is divided into a plurality of sound data, distributed to a plurality of servers, and transmitted to each server. It becomes difficult to reproduce the entire content from the content of the data, and the possibility that the content of the sound data is read by a third party can be further reduced.

According to the present invention, it is possible to provide a data conversion device capable of converting sound data into text data while achieving both convenience and security.

It is a figure which shows the network configuration of the data conversion apparatus which concerns on embodiment of this invention. It is a figure which shows the physical structure of the data conversion apparatus which concerns on this embodiment. It is a figure which shows the functional block of the data conversion apparatus which concerns on this embodiment. It is a figure which shows an example of the section of the sound data specified by the data conversion apparatus which concerns on this embodiment. It is a flowchart of the 1st process executed by the data conversion apparatus which concerns on this embodiment. It is a figure which shows an example of the minutes updated by the data conversion apparatus which concerns on this embodiment. It is a flowchart of the 2nd process executed by the data conversion apparatus which concerns on this embodiment. It is a figure which shows the other example of the section of the sound data specified by the data conversion apparatus which concerns on this embodiment. It is a figure which shows the example which specifies the section of the sound data by the data conversion apparatus which concerns on this embodiment.

An embodiment of the present invention will be described with reference to the accompanying drawings. In each figure, those with the same reference numerals have the same or similar configurations.

FIG. 1 is a diagram showing a network configuration of a data conversion device 10 according to an embodiment of the present invention. The data conversion device 10 cuts out sound data in a predetermined section from a series of sounds input by an input unit such as a microphone, and obtains the sound data in the section from the first voice recognition server 20 via the communication network N. , At least one of the second voice recognition server 30 and the third voice recognition server 40. The first voice recognition server 20, the second voice recognition server 30, and the third voice recognition server 40 are servers that generate text data based on the received sound data, and return the generated text data to the data conversion device 10. ..

Here, the communication network N is a wired or wireless communication network, and may be, for example, the Internet. The first voice recognition server 20, the second voice recognition server 30, and the third voice recognition server 40 are servers that provide a service for converting voice data into text data in the form of a so-called public cloud via the communication network N. It may be there. That is, the first voice recognition server 20, the second voice recognition server 30, and the third voice recognition server 40 may be servers that provide a service for converting voice data into text data without limiting users. In this example, a case where three servers are connected to the communication network N is shown, but the number of voice recognition servers that can be used as a public cloud is not limited to three, and is arbitrary. Further, the data conversion device 10 may be connected not only to a public cloud but also to a server that provides a service for converting voice data into text data in the form of a private cloud (that is, a form that limits users).

The data conversion device 10 cuts out, for example, sound data in a section including contents to be recorded as minutes from sound data recorded at a conference. The data conversion device 10 should not transmit the entire sound data recorded in the conference to the first speech recognition server 20, the second speech recognition server 30, and the third speech recognition server 40, but should record them as minutes. The sound data of the section including the content is cut out, and the sound data of the section is transmitted to at least one of the first voice recognition server 20, the second voice recognition server 30, and the third voice recognition server 40. In this way, by cutting out the sound data of the section to be converted into text and sending it to the server without sending the entire sound data recorded in the meeting to the server, a conversation that should not be leaked to the outside is performed in the meeting. Even if it is broken, the section to be converted into text data can be limited, and the sound data can be converted into text data while achieving both convenience and security of the cloud service.

FIG. 2 is a diagram showing a physical configuration of the data conversion device 10 according to the embodiment of the present invention. The data conversion device 10 includes a CPU (Central Processing Unit) 10a corresponding to a hardware processor, a RAM (Random Access Memory) 10b corresponding to a memory, a ROM (Read only Memory) 10c corresponding to the memory, and a communication unit 10d. And an input unit 10e and a display unit 10f. Each of these configurations is connected to each other via a bus so that data can be transmitted and received. In this example, the case where the data conversion device 10 is composed of one computer will be described, but the data conversion device 10 may be realized by using a plurality of computers.

The CPU 10a is a control unit that controls execution of a program stored in the RAM 10b or ROM 10c, calculates data, and processes data. The CPU 10a is an arithmetic unit that executes a program (data conversion program) related to control of conversion from sound data to text data. The CPU 10a receives various input data from the input unit 10e and the communication unit 10d, displays the calculation result of the input data on the display unit 10f, and stores the input data in the RAM 10b and the ROM 10c.

The RAM 10b is a storage unit capable of rewriting data, and is composed of, for example, a semiconductor storage element. The RAM 10b stores programs and data such as applications executed by the CPU 10a.

The ROM 10c is a storage unit capable of only reading data, and is composed of, for example, a semiconductor storage element. The ROM 10c stores programs and data such as firmware.

The communication unit 10d is an interface for connecting the data conversion device 10 to the communication network N, and is, for example, a LAN (Local Area Network), a WAN (Wide Area Network), the Internet, or the like configured by a wired or wireless line data transmission path. It is connected to the communication network N of.

The input unit 10e receives data input from the user, and includes, for example, a microphone, a keyboard, a mouse, and a touch panel.

The display unit 10f visually displays the calculation result by the CPU 10a, and is configured by, for example, an LCD (Liquid Crystal Display).

The data conversion program may be stored in a storage medium readable by a computer such as RAM 10b or ROM 10c and provided, or may be provided via a communication network connected by the communication unit 10d. In the data conversion device 10, the CPU 10a executes a data conversion program to realize various functions described with reference to the following figures. It should be noted that these physical configurations are examples and do not necessarily have to be independent configurations. For example, the data conversion device 10 may include an LSI (Large-Scale Integration) in which the CPU 10a and the RAM 10b or ROM 10c are integrated.

FIG. 3 is a diagram showing a functional block of the data conversion device 10 according to the present embodiment. The data conversion device 10 includes a determination unit 11, a sound data storage unit 12, a transmission unit 13, an extraction unit 14, a division unit 15, a reception unit 16, a synthesis unit 17, a correction unit 18, and a minutes storage unit 19. In this example, a case where these functional parts are realized by one computer will be described, but these functional parts may be realized by a plurality of computers.

The determination unit 11 determines whether or not a predetermined sound is included in the series of sounds input by the input unit 10e. Here, the predetermined sound may be any sound as long as it is a preset sound, and is, for example, a sound of ringing a physical bell or an electronically synthesized sound. You may do it. The determination unit 11 may be a trained model such as an RNN (Recurrent Neural Network) trained in advance so that a predetermined sound can be recognized. A user who uses the data conversion device 10 in a meeting can specify that a predetermined sound is played before a statement to be recorded in the minutes is made, and the content to be stated thereafter is converted into text.

Further, the predetermined sound may be a statement according to a predetermined rule. For example, the determination unit 11 may determine whether or not a predetermined sound is included depending on whether or not "Please minutes" is said after silence for about 2 seconds. Even in such a case, the determination unit 11 may be an RNN or the like learned in advance so that a predetermined sound can be recognized. The predetermined sound may be configured to be set for each user, and the predetermined sound can be set for each speaker or system user, and each user performs an operation of specifying the minutes target at an arbitrary timing. It may be a configuration that can be used. Also, in the example of inserting the text after the voice recognition result into the minutes to be described later, the minutes are memorized for each user, voice recognition is performed for the user who utters a predetermined sound, and the voice for the target user is performed. The recognition result may be inserted into the minutes for the target user, or the minutes for each user may be held individually. Then, after editing the minutes owned by only the person in charge of the minutes, it is also possible to upload the minutes to the file server for sharing.

The sound data storage unit 12 stores at least sound data in a section specified based on a predetermined sound in a series of sounds input by the input unit 10e. When the determination unit 11 determines that the series of sounds contains a predetermined sound, the sound data storage unit 12 may use at least a part of the series of sounds input after the predetermined sound as the predetermined sound. It may be stored as sound data of a section specified based on. For example, the sound data storage unit 12 may store sound data in a section that is input after a predetermined sound and until the predetermined sound is input again. In this case, the sound data storage unit 12 may store the sound data so that the predetermined sound is not included in the recording section. In this way, by storing at least a part of a series of sounds input after a predetermined sound as sound data in a section, the capacity of sound data to be stored in the storage unit can be reduced and stored. Since it is not necessary to determine after the fact whether or not a predetermined sound is included in the sound data, the calculation load can be reduced. The sound for specifying the start of the section and the sound for specifying the end of the section may be the same sound or different sounds. Further, the end of the section may be determined by the elapsed time from the time when a predetermined sound for specifying the start of the section is input.

Further, the sound data storage unit 12 may store a series of sound data input by the input unit 10e. The extraction unit 14 extracts at least a part of a series of sounds input after a predetermined sound from the series of sound data stored in the sound data storage unit 12, and the sound in a section specified based on the predetermined sound. Extract as data of. The extraction unit 14 may extract sound data in a section that is input after a predetermined sound and until the predetermined sound is input again. In this case, the extraction unit 14 may extract sound data so that the predetermined sound is not included in the extraction section. In this way, by storing the data of the input series of sounds and extracting at least a part of the series of sounds input after the predetermined sound, the data other than the sound data of the extracted section is also stored. It will be stored in the section, and data other than the sound data of the extracted section can be converted into text after the fact, and more flexible sound data can be converted into text.

When the determination unit 11 determines that a series of sounds contains a predetermined sound, the transmission unit 13 generates text data based on the sound data (first voice recognition server 20, second voice). The sound data of the section specified based on the predetermined sound is transmitted to the recognition server 30 and the third voice recognition server 40). Note that FIG. 3 does not show the server, but shows the communication network N.

The receiving unit 16 receives the text data generated based on the sound data of the section specified based on the predetermined sound from the server.

The division unit 15 divides sound data in a section specified based on a predetermined sound into a plurality of sound data. The transmission unit 13 may change the order of the plurality of sound data and transmit the plurality of sound data to the server. In this case, the receiving unit 16 receives a plurality of text data generated based on the plurality of sound data. Then, the synthesizing unit 17 synthesizes the received plurality of text data into one text data based on the rearrangement of the order of the plurality of sound data by the transmitting unit 13. In this way, the sound data in the section specified based on the predetermined sound is divided into a plurality of sound data, and the order is changed and transmitted to the server, so that the content of the transmitted sound data is third. It can be prevented from being read by a person.

Further, the transmission unit 13 may distribute and transmit the plurality of sound data obtained by the division unit 15 to a plurality of servers that generate text data based on the sound data. In the case of the present embodiment, the transmission unit 13 may distribute and transmit the divided plurality of sound data to the first voice recognition server 20, the second voice recognition server 30, and the third voice recognition server 40. By dividing the sound data of the section specified based on a predetermined sound into a plurality of sound data and distributing and transmitting the sound data to a plurality of servers, a part of the sound data transmitted to each server can be used as a whole. It becomes difficult to reproduce the contents of the sound data, and the possibility that the contents of the sound data are read by a third party can be further reduced. As a method of distributing the divided sound data to a plurality of servers, in addition to a method of randomly distributing the data, a configuration may be used in which the past voice data is recognized by voice and the distribution is preferentially distributed to a high-quality server. .. When preferentially distributing based on a high-quality server, the server may be determined and distributed for each user who is the speaker. Specifically, when the quality of the past voice recognition result of a certain user A is the best in the server α, the sound data related to the user A is preferentially distributed to the server α, and the sound data of the other user B is preferentially distributed. When the quality of the past voice recognition result is the best in the server β, the sound data related to the user B may be preferentially distributed to the server β.

The correction unit 18 corrects each word included in the obtained text data based on the reliability of the text conversion output by the server that executed the text conversion process. The first voice recognition server 20, the second voice recognition server 30, and the third voice recognition server 40 can convert sound data into text data for general terms, and are generally used for in-house terms and the like. If the sound data contains unspoken words, it may be difficult to correctly convert the sound data into text data. The correction unit 18 includes a language model such as RNN learned from the past minutes stored in the minutes storage unit 19 as learning data, and corrects words with low reliability in text conversion to correct words. conduct. As a result, even if the sound data contains words that are not commonly used, such as in-house terms, and it is difficult for the server to correctly convert the sound data into text, the minutes with more accurate content can be obtained. You will be able to create it.

The minutes storage unit 19 stores the text data obtained by converting the sound data in the form of minutes. The data conversion device 10 detects characters or symbols representing the agenda contained in the text data obtained by converting the sound data, and new text data is stored in an appropriate part of the minutes stored in the minutes storage unit 19. Is added. Further, the data conversion device 10 detects a character representing a person's name included in the text data obtained by converting the sound data, and assigns the name of the person in charge to an appropriate agenda of the minutes stored in the minutes storage unit 19. Append. These processes will be described in detail later with reference to FIG.

FIG. 4 is a diagram showing an example of a section of sound data specified by the data conversion device 10 according to the present embodiment. The figure shows the waveform of the sound data recorded at the conference. The sound data of this example includes the first section A1, the second section A2, and the third section A3. The first section A1 is a section corresponding to a statement that does not need to be recorded in the minutes, and the second section A2 is a section corresponding to a bell sound preset as a predetermined sound, and a third section. A3 is a section corresponding to a statement that needs to be recorded in the minutes.

The data conversion device 10 determines whether or not a predetermined sound is included in a series of input sounds by the determination unit 11. Here, whether or not a predetermined sound is included may be determined by whether or not a waveform of a predetermined sound is included. In this example, the determination unit 11 determines that a predetermined sound is included by inputting the waveform of the second section A2.

When the determination unit 11 determines that the series of sounds contains a predetermined sound, the sound data storage unit 12 uses the sound in the third section A3, which is input after the predetermined sound, as the sound based on the predetermined sound. It is stored as sound data of the specified section. After that, the data conversion device 10 transmits the sound data of the third section A3 to the first voice recognition server 20 and the like, and receives the text data corresponding to the content thereof.

Further, the data conversion device 10 divides the sound data of the third section A3 into a plurality of sound data and changes the order thereof before transmitting the sound data of the third section A3 to the first voice recognition server 20 or the like. It may be sent to a server or distributed to a plurality of servers. In this case, some division methods may be changed to adopt a division method in which the reliability of conversion to text data by the first speech recognition server 20 or the like is good. For example, a case where the sound data of the third section A3 is divided into three at equal intervals and the first voice recognition server 20 or the like recognizes each sound data, and a case where the sound data of the third section A3 is divided into 10 at equal intervals. The first voice recognition server 20 or the like may be made to recognize each sound data, and the reliability of text conversion may be compared, and a more reliable number of divisions may be adopted. As a result, the accuracy of voice recognition by the server can be improved, more accurate text conversion can be performed, and both the accuracy of text conversion and security can be achieved.

Further, the division point of the sound data may be determined according to the amplitude of the waveform. For example, the sound data may be divided at the center of a section where the amplitude of the sound data is equal to or less than a predetermined value. As a result, the voice recognition accuracy by the server can be improved and more accurate text conversion can be performed as compared with the case where the sound data is divided into a predetermined number of sections or a predetermined section width. , It is possible to achieve both accuracy and security of text conversion.

FIG. 5 is a flowchart of the first process executed by the data conversion device 10 according to the present embodiment. In the first process, when it is determined that the input series of sounds contains a predetermined sound, the sound data of the section specified based on the predetermined sound is stored, converted into text, and the minutes are recorded. Is the process of updating.

The data conversion device 10 acquires input sound data by the input unit 10e (S10). The determination unit 11 determines whether or not a predetermined sound is included in the input series of sounds (S11). When the input series of sounds does not include a predetermined sound (S11: No), the acquisition of sound data and the determination of whether or not the predetermined sound is included are continued.

On the other hand, when a predetermined series of sounds is included in the input series of sounds (S11: Yes), the sound data storage unit 12 stores the series of sounds input after the predetermined sounds based on the predetermined sounds. It is stored as sound data of the specified section (S12).

The division unit 15 divides the sound data of the specified section into a plurality of sound data (S13). The transmission unit 13 rearranges the order of the plurality of sound data and transmits the data to one or more of the first voice recognition server 20, the second voice recognition server 30, and the third voice recognition server 40 (S14). That is, the transmission unit 13 may change the order of the plurality of sound data and distribute the plurality of sound data to the plurality of servers for transmission.

The receiving unit 16 receives a plurality of text data obtained by converting a plurality of sound data into text from one or a plurality of servers (S15). The synthesizing unit 17 rearranges the order of the plurality of text data and synthesizes them into one text data based on the rearrangement of the order of the plurality of sound data and the distribution to the server (S16).

The correction unit 18 corrects a word having a low reliability into a word presumed to be appropriate based on the reliability of text conversion by one or a plurality of servers (S17). It is desirable that the correction unit 18 corrects the word after the synthesis unit 17 synthesizes one text data. This is because if a word is corrected in the state of a plurality of text data corresponding to a plurality of divided sound data, the context of the sentence becomes unclear and appropriate correction may be difficult.

The data conversion device 10 identifies a place to be added to the minutes based on a predetermined character in the obtained text data (S18). For example, it recognizes a character string or symbol representing a specific agenda, and adds the obtained text data to the part of the minutes where the agenda is described. Further, the data conversion device 10 identifies the person in charge based on the person's name in the obtained text data (S19). The data conversion device 10 may add the name of the identified person in charge to the minutes as the person in charge of the corresponding agenda.

Finally, the data conversion device 10 adds the date of the current day to the minutes as the description date, and updates the minutes (S20). In addition to the date, the time when the meeting was held may be added. This completes the first process.

FIG. 6 is a diagram showing an example of minutes D updated by the data conversion device 10 according to the present embodiment. Minutes D of this example include the content described on July 1st and July 3rd, and was updated on July 4th. Minutes D includes a description of the first agenda item D1 named "# 1000" and the second agenda item D2 entitled "# 2517 Verification of Certificate in Correct Way".

Regarding the first agenda item D1, the description includes "→ first described in the design document (described on 7/1)" and "→ discussed tomorrow (described on 7/3)". From this, it can be seen that as of July 3, it was decided to discuss the first agenda item D1 on July 4, the following day. Then, in the first agenda item D1, it is added that "→ ABC parameter is 1000 (described on 7/4)".

Such a description is added, for example, as follows. First, it is assumed that various discussions were held at the meeting and a conclusion was reached on what value the "ABC parameter" should be. At such a stage, when the bell corresponding to the predetermined sound is ringed, the data conversion device 10 outputs the sound data "Sharp 1000, ABC parameter is 1000", which is said after the predetermined sound is input. It is stored, transmitted to the first voice recognition server 20 or the like, and receives text data in which the contents are converted into text. Then, based on the character string "Sharp 1000", the text data "ABC parameter is 1000" is added to the description part of the first agenda item D1 named "# 1000". At this time, July 4 (7/4), which is the date of the day when the meeting was held, is added.

The second agenda item D2 is "→ 7/12 release schedule. The procedure has been linked to the application T. (7/1 description)", "→ [Mr. A's homework] Write in the release schedule of the quality control committee (7/3 description) ) ”,“ → Already described, please execute today. (7/4 postscript) ”is included.

Such a description is added, for example, as follows. First, it is assumed that various discussions were held at the meeting, and it was decided that it was necessary to carry out the task of "writing in the release schedule of the quality control committee", and it was decided that the person in charge would be "Mr. A". At such a stage, when the bell corresponding to the predetermined sound is ringed, the data conversion device 10 makes a statement after the predetermined sound is input, "Sharp 2517, Mr. A's homework, scheduled to be released by the quality control committee member. The sound data "write" is stored, transmitted to the first voice recognition server 20 or the like, and the text data in which the contents are converted into text is received. Then, based on the character string "Sharp 2517", the text data "Write in the release schedule of the quality control committee" is added to the description part of the second agenda item D2 named "# 2517". Also, based on the character string "Mr. A's homework", add "[Mr. A's homework]" to clarify the person in charge of the task. Then, July 3 (7/3), which is the date of the day when the meeting was held, is added.

As described above, according to the data conversion device 10 according to the present embodiment, it is possible to identify what kind of agenda the meeting is related to and add it to an appropriate part of the minutes. This reduces the workload of the minutes creator. In addition, the name of the person in charge can be identified and the person in charge can be added to an appropriate place in the minutes to support the smooth progress of the work.

FIG. 7 is a flowchart of the second process executed by the data conversion device 10 according to the present embodiment. The second process stores a series of input sounds, and when it is determined that the stored series of sounds contains a predetermined sound, the sound in the section specified based on the predetermined sound is recorded. It is a process to extract data, convert it into text, and update the minutes.

The data conversion device 10 stores the sound data input by the input unit 10e in the sound data storage unit 12 (S30). Here, the sound data may be continuously stored in the sound data storage unit 12 during the conference. The determination unit 11 determines whether or not a predetermined sound is included in the stored series of sounds (S31). When the series of sounds does not include a predetermined sound (S31: No), the second process ends.

On the other hand, when a predetermined sound is included in the series of sounds (S31: Yes), the extraction unit 14 sets the series of sounds input after the predetermined sound in a section specified based on the predetermined sound. It is extracted as sound data (S32).

The division unit 15 divides the sound data of the specified section into a plurality of sound data (S33). The transmission unit 13 rearranges the order of the plurality of sound data and transmits the data to one or more of the first voice recognition server 20, the second voice recognition server 30, and the third voice recognition server 40 (S34). That is, the transmission unit 13 may change the order of the plurality of sound data and distribute the plurality of sound data to the plurality of servers for transmission.

The receiving unit 16 receives a plurality of text data obtained by converting a plurality of sound data into text from one or a plurality of servers (S35). The synthesizing unit 17 rearranges the order of the plurality of text data and synthesizes them into one text data based on the rearrangement of the order of the plurality of sound data and the distribution to the server (S36).

The correction unit 18 corrects a word with low reliability to a word presumed to be appropriate based on the reliability of text conversion by one or a plurality of servers (S37).

The data conversion device 10 identifies a place to be added to the minutes based on a predetermined character in the obtained text data (S38). For example, it recognizes a character string or symbol representing a specific agenda, and adds the obtained text data to the part of the minutes where the agenda is described. Further, the data conversion device 10 identifies the person in charge based on the person's name in the obtained text data (S39). The data conversion device 10 may add the name of the identified person in charge to the minutes as the person in charge of the corresponding agenda.

Finally, the data conversion device 10 adds the date of the day as the description date to the minutes and updates the minutes (S40). In addition to the date, the time when the meeting was held may be added. This completes the second process.

FIG. 8 is a diagram showing another example of a section of sound data specified by the data conversion device 10 according to the present embodiment. The figure shows another example of the waveform of the sound data recorded at the conference. The sound data of this example includes a fifth section A5, a sixth section A6, a seventh section A7, and an eighth section A8. The fifth section A5 is a section corresponding to a statement that does not need to be recorded in the minutes, the sixth section A6 is an almost silent section for about 2 seconds, and the seventh section A7 is set as a predetermined statement. It is a section corresponding to the remark "Please give me the minutes", and the eighth section A8 is a section corresponding to the remark that needs to be recorded in the minutes.

The data conversion device 10 determines whether or not a predetermined sound is included in a series of input sounds by the determination unit 11. Here, whether or not a predetermined sound is included may be determined by whether or not a waveform of a predetermined sound is included. In this example, the determination unit 11 determines that a predetermined sound is included by inputting the waveforms of the sixth section A6 and the seventh section A7. That is, the determination unit 11 determines whether or not a predetermined sound is included in the series of sounds depending on whether or not "Please minutes" is said after silence for about 2 seconds.

When the determination unit 11 determines that the series of sounds includes a predetermined sound, the sound data storage unit 12 uses the sound in the eighth section A8 input after the predetermined sound as the sound based on the predetermined sound. It is stored as sound data of the specified section. After that, the data conversion device 10 transmits the sound data of the eighth section A8 to the first voice recognition server 20 and the like, and receives the text data corresponding to the contents thereof.

In this way, by determining whether or not a predetermined sound is included in a series of input sounds depending on whether or not a statement is made in accordance with a predetermined rule, a special sound (a special sound as a predetermined sound) is determined. For example, it becomes unnecessary to prepare for ringing a physical bell sound or an electronically synthesized bell sound, and it becomes possible to give an instruction to create minutes more easily.

FIG. 9 is a diagram showing an example in which a section of sound data is designated by the data conversion device 10 according to the present embodiment. FIG. 8 shows an example in which the 9th section A9 and the 10th section A10 are designated for the sound data including the 5th section A5, the 6th section A6, the 7th section A7, and the 8th section A8 shown in FIG. There is.

The data conversion device 10 displays the waveform of the recorded sound data and the recognized sound data section (in this example, the fifth section A5, the sixth section A6, the seventh section A7, and the eighth section A8). It may be displayed on 10f, and correction or addition of a section may be accepted from the user by a pointing device or the like included in the input unit 10e. For example, at the beginning of the meeting, I thought that it would be sufficient to record only the remarks made after generating a predetermined sound, that is, the remarks in the 8th section A8 in the minutes, but after the fact, the 5th section A5 You may want to keep some of the content discussed in the minutes. In such a case, the user can specify the section of the sound data to be extracted by the pointer PT or the like. In this example, the user designates the ninth section A9 and the tenth section A10 as a section to be newly extracted.

The data conversion device 10 transmits the sound data of the newly designated 9th section A9 and the 10th section A10 to the first voice recognition server 20 and the like, receives the text data in text form, and receives an appropriate part of the minutes. Add the text data to.

In this way, by displaying the section of the sound data to be converted into text so that it can be visually confirmed and making corrections and additions, it is possible to select the sound data to be converted into text more flexibly. Therefore, the convenience of the data conversion device 10 is improved.

The embodiments described above are for facilitating the understanding of the present invention, and are not for limiting the interpretation of the present invention. Each element included in the embodiment and its arrangement, material, condition, shape, size, and the like are not limited to those exemplified, and can be appropriately changed. Further, it is possible to partially replace or combine the configurations shown in the embodiments.

10 ... Data conversion device, 10a ... CPU, 10b ... RAM, 10c ... ROM, 10d ... Communication unit, 10e ... Input unit, 10f ... Display unit, 11 ... Judgment unit, 12 ... Sound data storage unit, 13 ... Transmission unit, 14 ... extraction unit, 15 ... division unit, 16 ... reception unit, 17 ... synthesis unit, 18 ... correction unit, 19 ... minutes storage unit, 20 ... first voice recognition server, 30 ... second voice recognition server, 40 ... Third voice recognition server, N ... Communication network

Claims (5)

A determination unit that determines whether or not a predetermined sound is included in a series of input sounds,
At least, a storage unit that stores sound data of a section specified based on the predetermined sound in the series of sounds, and a storage unit.
A division unit that divides the sound data in the section into a plurality of sound data,
When the determination unit determines that the series of sounds contains the predetermined sound , the order of the plurality of sound data is assigned to a plurality of servers that generate text data based on the sound data. A transmission unit that is interchanged and distributes and transmits each of the plurality of sound data to at least one server selected based on the information of the speaker of each of the plurality of sound data .
Receiving each of the plurality of text data generated based on each of the plurality of sound data from the at least one server, including the reliability of each word. Department and
A compositing unit that synthesizes the plurality of text data into one text data based on the rearrangement of the order of the plurality of sound data by the transmitting unit.
Based on a language model learned from past documents as learning data, a correction unit that corrects the word whose reliability is below a certain value, and a correction unit.
A data conversion device equipped with. When the determination unit determines that the series of sounds includes the predetermined sound, the storage unit may use at least a part of the series of sounds input after the predetermined sound in the section. Store as sound data,
The data conversion device according to claim 1. The storage unit stores the series of sound data and stores it.
Further provided is an extraction unit that extracts at least a part of a series of sounds input after the predetermined sound as sound data of the section from the data of the series of sounds stored in the storage unit.
The data conversion device according to claim 1. The correction unit is a word included in the one text data synthesized by the synthesis unit, and corrects the word whose reliability is a certain value or less.
The data conversion device according to any one of claims 1 to 3. The division unit divides the sound data in the section into the plurality of sound data based on the amplitude of the waveform of the sound data in the section.
The data conversion device according to any one of claims 1 to 4.

Images